I use this as an example of why we need to be careful to think about what we're optimising for in ML. They wanted to reduce the number of rats, but incentivised increasing the number of rat tails.
Do your performance measures correspond directly to things you care about? Are there ways of optimising the number while hurting something that matters to you? A good example to me is Google marking a picture of a black couple as "gorillas" - to a training algorithm that can easily be counted as a single wrong answer equivalent to any other when it is actually very significant.
Edit - this is the same for optimising any KPI, but ML is a case where your optimiser has no wider context and so there's more surprising cases. Worth thinking about overall though, what are you rewarding?
Something similar happened in Rio de Janeiro under the supervision of the epidemiologist Osvaldo Cruz during the very same year.
Rio was infested with bubonic plague and yellow fever. Oswald started a campaign to get the city rid of mosquitoes and rats including the same idea of buying rats from the population. The result were the same, people began importing rats from neighboring towns and breeding them at home. He later switched the focus into hygiene and sanitation.
> These days, the Great Hanoi Rat Massacre is mostly cited as an example of the “Cobra Effect,” an economic theory about how incentives, in a complex system, can lead to perverse, unintended consequences. Sometimes, it’s trotted out as argument against government intervention of any kind—but Vann says that that kind of misses the point.
> When they were first introduced, blue boxes on Canadian driveways and sidewalks seemed almost revolutionary. People were “extremely enthusiastic,” says Dan Hoornweg, a professor at the University of Ontario Institute of Technology. They “really wanted to recycle.” But the box came with conditions—an unfortunate compromise between environmentalists, government, and the soft-drink industry that created the perfect conditions for an explosion in litter.
If people are going to go ahead and cheat the system en masse, no system of incentives will ever work properly and they will languish.
Civil society requires us to generally do the right thing, most of the time, because there's ample opportunity to be shortsighted and cheat.
FYI Here's how Alberta git rid of rats [1], though I don't see any mention of rewards either, more of a publicly managed awareness program with hired pets killers.
I dont think anyone "cheated" in this case. Official government policy effectively turned rat tails into currency. Once that change was made, no amount of good intentions could prevent the inevitable.
You end up hiring "professionals" that will actually keep the spirit of the agreement not just the letter (contracts do wonderful things in the modern world).
Although, I do know of one case where the locals gave up on a building (exterminator said no, the cats ran for it, and hantavirus was a concern). They cut a trench around the building, filled it with gas, lit the trench up, then set the building on fire. It was a bit horrific. They built another building on the same plot but not the exact same place.
Reminds me of that hoarders house in the US, that was so infested with cockroaches the fire department built a fire pit around it and burned it to the ground.
Its a valid strategy, especially when you are dealing with disease or some insects. It provides a bit of a training opportunity and sometimes there just isn't any hope.
If for some odd reason, you are doing thing yourself (why?!), do remember the order: dig trench -> fill trench -> ignite trench -> ignite structure. Do the order of the last two step wrong can result in a very scary video on YouTube. Do not have squeamish people there with you. Keep plenty of firefighting gear near.
“To watch out for programs being created in situations where where the arrogance is so strong and the power differential is so intense that evidence can be ignored.”
I use this as an example of why we need to be careful to think about what we're optimising for in ML. They wanted to reduce the number of rats, but incentivised increasing the number of rat tails.
Do your performance measures correspond directly to things you care about? Are there ways of optimising the number while hurting something that matters to you? A good example to me is Google marking a picture of a black couple as "gorillas" - to a training algorithm that can easily be counted as a single wrong answer equivalent to any other when it is actually very significant.
Edit - this is the same for optimising any KPI, but ML is a case where your optimiser has no wider context and so there's more surprising cases. Worth thinking about overall though, what are you rewarding?
See also: https://wiki.lesswrong.com/wiki/Paperclip_maximizer
Something similar happened in Rio de Janeiro under the supervision of the epidemiologist Osvaldo Cruz during the very same year.
Rio was infested with bubonic plague and yellow fever. Oswald started a campaign to get the city rid of mosquitoes and rats including the same idea of buying rats from the population. The result were the same, people began importing rats from neighboring towns and breeding them at home. He later switched the focus into hygiene and sanitation.
Cruz is also famous for making vaccination mandatory in Brazil, a decision that led to the infamous Vaccine Rebelion in Brazil: https://en.wikipedia.org/wiki/Vaccine_Revolt
The general name for this phenomenon is the Cobra effect: https://en.wikipedia.org/wiki/Cobra_effect
From the article:
> These days, the Great Hanoi Rat Massacre is mostly cited as an example of the “Cobra Effect,” an economic theory about how incentives, in a complex system, can lead to perverse, unintended consequences. Sometimes, it’s trotted out as argument against government intervention of any kind—but Vann says that that kind of misses the point.
The cobra effect seems to be the theme of the day.
From the other article on the front page today, "Why Recycling Doesn't Work". https://news.ycombinator.com/item?id=19889365
> When they were first introduced, blue boxes on Canadian driveways and sidewalks seemed almost revolutionary. People were “extremely enthusiastic,” says Dan Hoornweg, a professor at the University of Ontario Institute of Technology. They “really wanted to recycle.” But the box came with conditions—an unfortunate compromise between environmentalists, government, and the soft-drink industry that created the perfect conditions for an explosion in litter.
Michael Vann wrote this recent book on the Hanoi Rat Hunt
https://global.oup.com/ushe/product/the-great-hanoi-rat-hunt...
The moral of the story is wrong.
If people are going to go ahead and cheat the system en masse, no system of incentives will ever work properly and they will languish.
Civil society requires us to generally do the right thing, most of the time, because there's ample opportunity to be shortsighted and cheat.
FYI Here's how Alberta git rid of rats [1], though I don't see any mention of rewards either, more of a publicly managed awareness program with hired pets killers.
[1] https://www.alberta.ca/history-of-rat-control-in-alberta.asp...
I dont think anyone "cheated" in this case. Official government policy effectively turned rat tails into currency. Once that change was made, no amount of good intentions could prevent the inevitable.
Alberta really hates rats.
I am interested in knowing what happened after. Did they just give up on the rats? What do cities around the world do to control the rat population?
You end up hiring "professionals" that will actually keep the spirit of the agreement not just the letter (contracts do wonderful things in the modern world).
Although, I do know of one case where the locals gave up on a building (exterminator said no, the cats ran for it, and hantavirus was a concern). They cut a trench around the building, filled it with gas, lit the trench up, then set the building on fire. It was a bit horrific. They built another building on the same plot but not the exact same place.
Reminds me of that hoarders house in the US, that was so infested with cockroaches the fire department built a fire pit around it and burned it to the ground.
Its a valid strategy, especially when you are dealing with disease or some insects. It provides a bit of a training opportunity and sometimes there just isn't any hope.
If for some odd reason, you are doing thing yourself (why?!), do remember the order: dig trench -> fill trench -> ignite trench -> ignite structure. Do the order of the last two step wrong can result in a very scary video on YouTube. Do not have squeamish people there with you. Keep plenty of firefighting gear near.
I don't know the history, but the endgame modern consensus is that decreasing garbage availability is the only way to go.
Interesting! Whatever you optimize for in human (or machine) behavior is what you get.
“To watch out for programs being created in situations where where the arrogance is so strong and the power differential is so intense that evidence can be ignored.”
~historian Michael Vann
TL;DR: rat hunters were cutting rat's tails to obtain the reward, leaving rats alive to let them multiply.